Response-to-repeatability of quantitative imaging features for longitudinal response assessment

Area of Science:

• Medical Imaging
• Radiomics
• Oncology

Background:

• Quantitative imaging biomarkers (QIBs) are crucial for treatment response assessment.
• Repeatability is a common metric for selecting QIBs, but sensitivity to change is also vital.
• Existing metrics do not fully capture a QIB's utility in measuring treatment effects.

Purpose of the Study:

• Introduce the response-to-repeatability ratio (R/R) to evaluate QIB sensitivity to treatment-induced changes.
• Apply R/R to PET texture features in bone-metastatic prostate cancer.
• Compare R/R with traditional repeatability metrics like ICC and CV.

Main Methods:

• Calculated R/R as the proportion of follow-up changes outside the 95% limits of agreement (LOA) derived from test-retest data.
• Analyzed 47 texture features from 18F-NaF PET/CT scans in prostate cancer patients.
• Assessed intraclass correlation coefficient (ICC) and coefficient of variation (CV) for each feature.
• Correlated R/R with ICC and CV using Spearman's rank correlation.

Main Results:

• R/R varied significantly across texture features, with 87% showing R/R > 5% and 23% showing R/R > 20%.
• Repeatability metrics (ICC, CV) showed weak correlations with R/R (ρ=0.40 and ρ=0.23, respectively), indicating limited ability to predict sensitivity to change.
• Skewness, kurtosis, and diagonal moment demonstrated high ICC (>0.75) and superior R/R values compared to SUVmax.

Conclusions:

• The R/R metric effectively characterizes the sensitivity of QIBs to detect measurable changes during treatment.
• R/R complements existing precision metrics (CV, ICC, LOA) for assessing QIB utility in treatment response evaluation.
• This approach aids in selecting robust QIBs that are both repeatable and sensitive to therapeutic effects.